Amazon Will Let Claude Run Your Store. It Will Not Let Claude Buy Anything From One.
Amazon opened Seller Central's APIs to outside AI agents like Claude this week — the same week it blocked Meta's shopping agent, Muse, from buying anything on Amazon at all.
At its Accelerate seller conference in Seattle this week, Amazon announced that independent sellers can now run their entire business through an outside AI agent — specifically Anthropic’s Claude, or Amazon’s own Quick assistant — without ever logging into Seller Central. Connect Claude to your seller account, a process Amazon says takes about sixty seconds and no code, and it can check inventory, adjust prices, update listings, and read your sales analytics, asking your approval before it takes any action. “Our vision was that they would never have to log into Seller Central,” said Mary Beth Westmoreland, Amazon’s VP of Worldwide Selling Partner Experience, describing a future in which the interface Amazon spent two decades building becomes optional, so long as the thing replacing it is an AI agent Amazon has personally authorized.
That’s the operative word: authorized. Because in the same week, on the other side of the same website, Amazon flatly blocked a different AI agent from doing anything at all.
Meta launched Muse, its own personal shopping agent, on September 8. Within two weeks, Amazon had cut it off from Amazon.com entirely, greeting it with a message reading “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use.” Amazon’s complaints were specific: Muse never told Amazon it would be browsing the site on customers’ behalf, it didn’t identify itself as automated while doing so, and — the part that reads worst in a press release — it was capturing and storing customer login credentials while quietly handling transactions and account data Amazon says it never agreed to expose. “Third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly,” Amazon said, in a statement that is either a reasonable security standard or a extremely convenient one, depending on whose agent is asking.
It’s not a new fight for Amazon. The company has sued Perplexity over its Comet browser and has previously blocked shopping agents built by Google and OpenAI from operating on its site, all on broadly similar grounds — undisclosed automation, unauthorized credential handling, an agent acting as a customer’s proxy without telling the store it’s doing so. Meta and Amazon are, notably, still business partners in other respects: Amazon products have been purchasable through Facebook and Instagram since 2023, and Meta signed a multibillion-dollar chip deal with Amazon back in April. None of that stopped Amazon from locking Muse out the moment it tried to buy something without asking first.
Put the two announcements side by side and the shape of Amazon’s actual policy comes into focus, and it isn’t really about agents at all — it’s about who’s driving. An AI agent that helps a merchant sell things on Amazon, using an integration Amazon built, reviewed, and can revoke, gets a stage at a conference and a quote from a VP. An AI agent that helps a customer buy things from Amazon, using an integration Amazon didn’t build and can’t fully see inside of, gets a Conditions of Use violation notice and a lawsuit-shaped glare. Both are, technically, “AI agents transacting on Amazon.” Only one of them cuts Amazon out of deciding how.
This is, in fairness, a coherent position and not merely a self-serving one — the credential-capture and undisclosed-automation complaints against Muse are real security concerns, not manufactured ones, and a marketplace has a legitimate interest in knowing whether the “customer” at checkout is a person or someone else’s unaudited AI wearing a person’s login. But it’s also a position that happens to land, with suspicious precision, on “agents Amazon owns the pipe for: welcome. Agents Amazon doesn’t: get out.” Sellers get a friendly, sanctioned Claude integration with guardrails Amazon itself designed. Meta gets a 403.
The Seller Assistant plugin is currently in beta for U.S. stores, with the promise of international rollout to come, and Amazon is sweetening the deal with a free year of Quick Plus for every primary account holder through the end of 2026 — an incentive structure that suggests Amazon would very much like sellers using its approved rails before anyone gets any ideas about building their own. Meanwhile Muse remains locked out, Meta hasn’t said publicly how it plans to respond, and the era of AI agents doing your shopping for you is arriving less as a single unified marketplace and more as a patchwork of walled gardens, each one deciding, store by store, which robots it trusts to hold the keys — and which ones it’s just going to trust for now, from the seller side, thank you very much.
Sources:
- Amazon opens its seller tools to outside AI agents, starting with Anthropic’s Claude — GeekWire
- Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic’s Claude — About Amazon
- Amazon’s new plugin lets sellers manage their entire e-commerce business with AI — SiliconANGLE
- Amazon blocks Meta’s Muse AI assistant in new standoff over agentic shopping — GeekWire